Currently Being Moderated
wjpeter

Criteria for HPC Storage are evolving

Posted by wjpeter in Big Data Bingo on Jun 20, 2011 2:51:11 PM

By Ingo Fuchs

 

In the world of High-Performance Computing - HPC - the technology leapfrogging is unparalleled. The demands on computing environments and the storage infrastructure supporting them are immense and ever-changing. Interestingly, in my conversations with customers in this space the storage conversation often boils down to four key criteria:

 

-          Density, either in terms of performance or capacity that can be accomplished per datacenter floor tile

-          Higher and higher performance requirements from ever growing data intensive workloads

-          Capacity, with the goal of processing  as much as data as possible

-          Power, Cooling and Weight, where the best storage system can’t win if it doesn’t fit the datacenter specs

 

As many of the world’s leading HPC experts again gather at ISC in Hamburg, Germany from June 19th-23rd 2011 it is interesting (and satisfying) to see that storage has become increasingly a focus in “Advancing the frontiers of supercomputing”, ISC’s motto this year. NetApp is there, of course, and will be talking about a number of interesting developments.

 

First – NetApp teamed up with Whamcloud to benchmark the NetApp E5400 with the Lustre file system on the Lawrence Livermore National Labs Hyperion test bed. In only 4U, the test shows 5GB/s read and 3.5GB/s writes, thus enabling a standard 40U rack deployment to achieve up to 50GB/s read throughput per rack. This is proof that a storage system with high density and low power requirements can easily deliver outstanding real-world-ish performance results as well.

 

Second – Bull, a global OEM provider based in Europe, is building OEM solutions based on the E5400, with customers like CEA in France already using it. One thing that stands out to me in the implementation at CEA is that “zero downtime” is becoming a common requirement in the HPC space, where only a few years ago pure performance was everything.

 

So, I will amend my list above to add:

 

-          Uncompromised uptime, delivering predictable performance, density AND availability

 

It’s an exciting field and a great show, make sure to check out the NetApp solutions for HPC if you are active in this area. Do you have any comments? Any other criteria to add to my list?

 

Thanks for reading,

 

Ingo

Comments

Filter Blog

By author: By date:
By tag: